Office 365 Distilled

The boys put on their harness and see how they can govern AI. 

Creators and Guests

Host
Marijn Somers
Microsoft MVP. Freelance Microsoft 365 expert focusing on user adoption and governance. Trainer. Licensed watchmaker.
Host
Steve Dalby
Podcaster "Office365Distilled" Driving Collaboration Business Goals, Speaking about Governance, Whiskey taster and imbiber all round father and good guy.

What is Office 365 Distilled?

This podcast is presented by Marijn and Steve; two experienced Microsoft 365 and SharePoint consultants. In each podcast you will get a smile or two with a chunk of interesting and practical help in Microsoft 365. Then you get introduced you to what we hope is a Whisky you have not tasted before.

Speaker 1:

Hey. Hey. So just like a cart horse, we're gonna talk about harnesses.

Speaker 2:

Cart horse. Yes. That that was not the angle I wanted to go in, but... Yes. You said, do you know

Speaker 1:

what a harness is? And I went, yes. I do know what a harness is. He went, not that kind of harness.

Speaker 2:

I know where your mind wanders off to.

Speaker 1:

I know. I know. I just like to try and surprise you with it.

Speaker 2:

Yeah. Well, there you are.

Speaker 1:

All part of the fun.

Speaker 2:

Exactly. You're

Speaker 1:

plugged in properly at the bottom, I'm guessing. Yep. Good. I'm gonna stop worrying about it.

Speaker 2:

Yeah. So we just started recording and we saw something on the screen that my mic was not really

Speaker 1:

Doing what it's supposed to be.

Speaker 2:

Up a lot of sound, but playback worked. Yeah, we'll see.

Speaker 1:

I can always fix it afterwards. We know that you're in there somewhere. Exactly. Just that I'm bigger than you.

Speaker 2:

And that's fine. I don't have a problem with that. Yeah. And I mean, otherwise, I... You can just AI me in.

Speaker 2:

So I mean... Yeah.

Speaker 1:

True. I

Speaker 2:

give you my harness and then you can... Then

Speaker 1:

lead you along the way.

Speaker 2:

Yes. Thought, no. Oh, no. Horrible.

Speaker 1:

Do you know I saw that once ago in there? So, Dimitri and I went out for lunch when I was working at Euroclear and you wander down the ladies, it's a nice sunny day, you wander past the ladies windows just for fun and then wander back again just to get some physical exercise.

Speaker 2:

Yeah yeah yeah.

Speaker 1:

And there was a young gentleman. Well, I don't know how young he was, but there was a gentleman on his hands and knees with a lead around his neck with a black mask and a ball gagging Okay. Being led around the feet of the ladies in the the window. So it's unusual,

Speaker 2:

you know? Well, some call it unusual, other people call it Friday. I mean, I get it.

Speaker 1:

Yeah. That is true. But the funniest thing was people were going, what the fuck? And they were stopping to look, and of course the girls were trying to

Speaker 2:

go, go away, leave it alone.

Speaker 1:

You know, look, the guy's decided he wants to dress up with a lead on and go in the front window. We're gonna be entertained and have fun. It was just strange.

Speaker 2:

Of course.

Speaker 1:

Yeah. So there was... So... But it's not that kind of harness. No.

Speaker 1:

Apparently not. Alright. We'll come to that in a little while.

Speaker 2:

Speaking gigs? Yes. Gdansk Baltic Summit is coming up in a few weeks, two weeks.

Speaker 1:

Seventeenth, eighteenth or something?

Speaker 2:

'24, '25.

Speaker 1:

Oh, that's right. Yes. '26,

Speaker 2:

something like that.

Speaker 1:

So I think I'm doing Bletchley.

Speaker 2:

Mhmm.

Speaker 1:

Okay. Then flying to Gdansk. To Zagreb. Zagreb. Yeah.

Speaker 1:

You've got me going now. You said Zagreb, and now I said Gdansk, but reversing everybody. Okay. So, yeah, to Zagreb. Yeah.

Speaker 1:

So I think that's a Friday. So Wednesday at Bledchley, and then Friday in Gdansk. Yeah. Sorry. Fuck it, hell.

Speaker 1:

Listen. As long as I get on the right plane, it does not matter. Exactly. It does not matter.

Speaker 2:

And then Scottish Summit. And then we're going to Scottish Summit. Yes. We're be Very cool. Some live streaming.

Speaker 2:

I have a session as well. Yep.

Speaker 1:

Yeah. Nook's got a workshop.

Speaker 2:

Yes. With Sandra and

Speaker 1:

Jay Jay somebody? Jay Jay. It's an interesting session. I will be busy, so we don't wanna go, but, yeah, they're gonna do gamification of workflows and automation. Yeah.

Speaker 1:

So I think one of them is doing AI, one's doing gamification, one's doing power apps. So that's a whole day workshop.

Speaker 2:

I already know who does woo, but...

Speaker 1:

Yeah. Know. So do I.

Speaker 2:

Yeah. Yeah. Okay.

Speaker 1:

So that was that was, that was, interesting. So that would be pretty cool. Nice. We had did some video recording for Scottish Summit this last couple of weeks. Yep.

Speaker 1:

Very interesting section with Rich Rob. Rich Rob. Yeah. But anyway, that that'll be coming out on the Scottish Summit's socials, but, yeah, they're doing a train the trainer or a speaker session, first speaker session, which was Oh, Rex. Rex.

Speaker 2:

Yes. Rex. Yeah.

Speaker 1:

Yeah. Yeah. So so, yeah, that's all coming together. Are you going to the cabaret on the Saturday?

Speaker 2:

Did we need For dinner. To book tickets?

Speaker 1:

Or Yeah. That's why I'm saying it. There's a discount code. We get Oops. Speaker money off it.

Speaker 1:

We still got time.

Speaker 2:

Okay. Because But it's like

Speaker 1:

we did the James Bond thing.

Speaker 2:

No. I did because Sarah Fena emailed me about this

Speaker 1:

No. She was emailing about the speaker's dinner.

Speaker 2:

Oh, that's something else. Okay.

Speaker 1:

Do you know what? I thought I was the one that's got no fucking admin or sense of diary.

Speaker 2:

Jeez. Okay. No. Okay.

Speaker 1:

I'll You're still on holiday, aren't you? It's only the September 1.

Speaker 2:

Yeah. Exactly. So I, yeah, I I need to double check that. I've I've got no idea. Okay.

Speaker 1:

Yeah. Yeah. So that's the Saturday. It's the same thing we did before.

Speaker 2:

So Oh, it's that casino night Yeah. Kind thing. Awards thing.

Speaker 1:

Yeah. And we got the awards. Are we gonna get people to vote for us on the awards?

Speaker 2:

Should we?

Speaker 1:

I don't know. We never win.

Speaker 2:

Nobody Yeah. Loves And is is there like a podcast category?

Speaker 1:

Oh, yeah. Of course, there is.

Speaker 2:

So, yeah, so it's probably gonna go to the Zoe and CoCo. CoCo, Co Pilots, CoCo Nation.

Speaker 1:

CoCo Pro, like, yeah. Well, there was some talk about maybe people shouldn't be able to win it twice, but who knows what will happen.

Speaker 2:

This bullshit is like Oscars. I mean Yes. You you can win it more than once. That's fine. They're delivering really good stuff.

Speaker 2:

They're also talking about harnesses, by the way. Are they? Yes. Because it was So we and Kevin and the harnesses. Don't go there.

Speaker 1:

It's too late. Too late. I know who the dominatrix would be.

Speaker 2:

Oh, for sure. I mean, that's an easy one. I know who will be crawling around on hands and knees with a ball gag. Yep. That's an easy one.

Speaker 1:

Sorry, Kevin. Don't throw my whiskey away because he's actually holding a bottle of whiskey for me at the moment. But no, that was was slipped out. I couldn't resist. But no, I know they do very well.

Speaker 1:

I know people speak Oh, highly yeah. Of Absolutely. Yeah. But they're serious. We're not serious.

Speaker 2:

No. Exactly. And that's why people love us.

Speaker 1:

Yeah. I think so. Yeah. I think so. So, yeah, we got a busy few days afterwards because we finally get to go and see the casks.

Speaker 2:

Yes. We're gonna visit our own casks. Well... Yeah.

Speaker 1:

Well, the ones

Speaker 2:

that we bought, not the ones that we build or whatever, but

Speaker 1:

Distilled. People

Speaker 2:

bought it.

Speaker 1:

Yeah. Yes. And then we

Speaker 2:

just see what we wanna do with it.

Speaker 1:

Yes. What was yours? Like, I was trying to remember. Lerlincourt. Oh, wow.

Speaker 1:

That should

Speaker 2:

be nice. Five year old, I think.

Speaker 1:

Five year be the same age as mine then. Mine's five. Yeah. So my Ana... Ananise or Annapolis or whatever it's called.

Speaker 2:

Yeah. Ard... Ardmore?

Speaker 1:

No. The Ardmore I've got.

Speaker 2:

Yeah. The Ardmore and the Annandale? Annandale. Yes.

Speaker 1:

That was it. Yes. So the Annandale, we've tasted. Yeah. That was

Speaker 2:

rough, but that was

Speaker 1:

going in

Speaker 2:

nice direction.

Speaker 1:

Yeah. So we'll see what that is, but the, Admoor is the killer. So we have to decide whether we're gonna bottle it or not. Uh-huh. Uh-huh.

Speaker 1:

Or move it to an interesting cask.

Speaker 2:

Uh-huh.

Speaker 1:

Indeed. Tempted to do that just for the hell of it.

Speaker 2:

Yeah. But, I mean, it's all been done now. I mean, Madeira cask, red wine cask, white wine cask, rum cask, tequila, mezcal. Tequila. I mean, you habanero cask.

Speaker 2:

Let's do something spicy. Spicy whiskey.

Speaker 1:

Oh, dear. Anyway, so, yeah, we're gonna do that afterwards, so that'll be fun. And we have a sponsor coming up. It won't be the next episode. It will be one afterwards.

Speaker 1:

Yes. That looks really good. Yes. The wiki, the power wiki, what do you call it? Oh.

Speaker 1:

Wait. This is not the real

Speaker 2:

I app looked it up because I went to their website because I was like, what is this? And then I went through the website and I was like, oh, this is freaking cool.

Speaker 1:

I know it is. Yeah. So Wiki's a boring things. Alright? I mean, ever since SharePoint came along and got itself together with its pages and stuff Mhmm.

Speaker 1:

And with all of its security and governance, then that's the way to go. You know, Wiki's kind of past it. Yeah. But they've integrated AI into Wiki, which has simplified the whole process. It's really very cool.

Speaker 2:

Yeah. It it it kinda looked like Viva Engage on steroids. Yeah. Yeah. It did.

Speaker 2:

Yeah. So I can ask you questions, and it will either find the answer itself or let someone else reply to it.

Speaker 1:

We'll probably fill it all out.

Speaker 2:

Who's pretty cool.

Speaker 1:

Yeah. Yeah. So that's gonna be good. So, Petro... Petro...

Speaker 1:

No. Il e m. Yes. I l a

Speaker 2:

I just

Speaker 1:

skimmed through

Speaker 2:

the email. I quickly went to the website in between two meetings.

Speaker 1:

But that... So next time, we'll give you all the details. And then the, podcast afterwards, we will taste, a really nice amarut.

Speaker 2:

Yes.

Speaker 1:

From India. A really, really nice one, which is I am looking forward to fees for the sponsorship. Yeah. And then we'll interview him online and we'll Yeah. We'll see how that goes out.

Speaker 1:

So that should be fun. So we've got a bit of sponsorship. Mhmm. And I guess you put it... You're doing, no.

Speaker 1:

You did Experts Live, didn't you? No?

Speaker 2:

In June. Yes.

Speaker 1:

Yeah. Yeah.

Speaker 2:

You did that. The UK one. Yeah.

Speaker 1:

Yeah. And I've put in for a number of other episodes this year. So interesting one in Casablanca, because I've never been to Casablanca, so I'll put in

Speaker 2:

for it.

Speaker 1:

It was an AI one, I think. Or I

Speaker 2:

I I went to Casablanca a few years ago.

Speaker 1:

Oh, okay.

Speaker 2:

But don't get your hopes up too much. It's it's it's really like an industrial workers' town.

Speaker 1:

Oh, the city itself?

Speaker 2:

Yeah. City itself. Yeah. It's got a really freaking big, huge mosque, but other than that, it's pretty

Speaker 1:

Oh, I see.

Speaker 2:

It's it's not that pretty. Yeah.

Speaker 1:

I will. Anyway, we'll see whether that happens or not. Yeah. But there we go. Alright.

Speaker 1:

So, yeah, so that's our calendar for the next few weeks. Speaking has started yet again.

Speaker 2:

Yeah. But I mean, I've only got, well, I've got these two. I've got Gdansk, the Baltic Summit, and then I've got Scottish. And then there's the Belgium one.

Speaker 1:

Oh, Belgium one. Yes.

Speaker 2:

You mean Belgium one. But then I don't have anything else planned for the end of the year,

Speaker 1:

so I'm...

Speaker 2:

I am gonna submit something for Oslo. Yep. Yeah. Exactly.

Speaker 1:

So... Yeah. Well, we'll see. Mean, to be to be honest, it's quite nice because they got so many of them before the summer. They were just...

Speaker 1:

We're tripping up all over the place.

Speaker 2:

Yeah. I think I already did, like, 12 conferences by now. So... Yeah. But still, I mean, it it it feels empty.

Speaker 2:

It feels a little weird.

Speaker 1:

Having to twist your thumbs at home and not, do this stuff.

Speaker 2:

Yeah. Exactly. Like, there's no no other things to look forward to.

Speaker 1:

Yeah. I got it. I got it. So which girl was it that asked you to, take part in a harness or

Speaker 2:

super technical, she is. So I kinda said like, oh, do you wanna do a session together? Something maybe AI and SharePoint or whatever. And then she came back, but something about... Yeah.

Speaker 2:

Something about the SharePoint list forms and the Copilot Studio new Harness. And I was like, I don't understand any... The only thing I understand is Copilot list form... SharePoint list form. That's the only thing I I know about.

Speaker 2:

Harness? Did you tell

Speaker 1:

her that? So she doesn't know you don't know yet.

Speaker 2:

No. No. I told her. Right? It's just like, I have no idea what this is.

Speaker 2:

So I had to look it up, and now the funny thing is everywhere is the word harness.

Speaker 1:

I

Speaker 2:

mean, Zoe and Kevin just recorded a new episode about Copilot Harness. I saw it on another one. I think it was the... Not the guardians of the governance, but, well, one of the other podcasts. There's a, an an article, news article from Microsoft that I was reading that was talking about the harness.

Speaker 2:

It's like, it's everywhere. Suddenly.

Speaker 1:

Well, was... Let's let's... I mean, I did some research, of course, because I had no idea what it was either. I knew what it was, but apparently we were talking about different harnesses. So it appears that it became popular and crystallized as the term was harness engineering in February of this year, alright, around AI.

Speaker 1:

So it's all about the environment surrounding an AI agent, and I'll dig into it in a in a little while. So it... Rather than just, you know, improving prompts, it's about all the things that make something work. And then apparently, OpenAI published Harness Engineering Leveraging Codecs in an Agent Led World on the February 11. And if you think about it, you kind of, March, April, May, summer holidays.

Speaker 1:

So it's probably seeped through and down and into everybody's reading list. Yeah. Maybe. So, basically, they form... What they're saying is agent equals model plus harness.

Speaker 1:

So that's where it came from. The best definition that I came out of... Oh, by the way, just just so I can name drop, because I know this off by heart. Mitchell Hashimoto was the co founder of HashiCorp and creator of Terraform. He was the guy that did the name drop.

Speaker 1:

Gotcha. But apparently, it's not a new name, like all things. It's been around for decades, apparently. So anyway, it's just... It's bigger or popular.

Speaker 1:

But the definition was rather nine. So my AI says, if you mean AI harness, it's basically the control layer around an AI model or agent.

Speaker 2:

Mhmm.

Speaker 1:

So whenever we put an agent together, it's just just the name and the the task of automating it. It needs a language model. It needs some orchestration. It needs some APIs. It needs some...

Speaker 2:

So it's

Speaker 1:

the connection of those. And and again, my AI said, the AI in Harness is everything that makes that clever bit useful and safe in the real world. So it's the instructions, the tools, the rag, the knowledge sources, permissions, workflows, memory, validation, login security, and all the rules that we may or may not have around AI. So that's what it is. So basically, it's a little bit like your content type.

Speaker 1:

That's a nice analogy. Content type has the document, the metadata, potentially some

Speaker 2:

It's your it's your document library Yeah. Is is the model and then everything around it, the security model, the information, architecture, yeah, metadata, that's all your Yeah.

Speaker 1:

So basically, if you're documenting your AI, you would document everything within the boundary of that harness.

Speaker 2:

But why why not just call it governance? Like, we all called everything governance for the last Because governance

Speaker 1:

is about the rules and the guidelines. That's true. It's not about the workflows or the content attached to it. So you need a neutral I

Speaker 2:

don't know why they went with the word harness. I don't know. Is it because it's it's holding

Speaker 1:

the model together and... You know, somebody came up with the word penis. Alright. Somebody somewhere thousands of years ago said Yeah. Hey, that's penis.

Speaker 2:

No. It's because the pen is from your brother. That's that's the one. The pen is. The pen is from your brother, and that's what...

Speaker 2:

That's how it how it got there.

Speaker 1:

But all I'm trying to say in a very rude way is that, you know, somebody comes up with something and it catches on and

Speaker 2:

and they leave Okay. But why harness? Why not, the headphones of AI or the air conditioning something. Or

Speaker 1:

Harness is for containing things and controlling things.

Speaker 2:

Mean, it's like a medieval harness that holds the model in place or something.

Speaker 1:

Yes. A harness that... It's the reins of a horse would be a harness.

Speaker 2:

Yes.

Speaker 1:

It's the netting on the back of a lorry would be a harness

Speaker 2:

Oh, that

Speaker 1:

holds all the products in. Yeah. So by... I can understand why they logically go as a harness. It's what's contained.

Speaker 1:

It's keeping everything contained Yeah. Okay. And under control Okay. And tied

Speaker 2:

Okay.

Speaker 1:

So can you not remember the last time you were tied down?

Speaker 2:

I I I can't even remember what I did yesterday.

Speaker 1:

So No. Okay. I get that. So... But that would be the harness.

Speaker 1:

Yeah. Okay. So it kinda makes sense. But the problem is that it's not distinctive. It's just a vague collective noun for everything associated with a particular agent.

Speaker 2:

It's it's a vague it's a vague term as well because it's just like you got the model and then this is everything around the model.

Speaker 1:

Okay. And it might be one thing. It might be 10 things. Yeah. So it's it's difficult.

Speaker 1:

I wonder if I look at my chat GPTs, some of them are workspaces. Of them are just your

Speaker 2:

agents or your Yeah. Just my chats.

Speaker 1:

Yeah. Chat GPTs. Some of them are agents, So I have one that goes off to my email every Friday and just make sure that I don't have any emails about bills that I've not paid or anything. Yeah. I have...

Speaker 1:

But I have a connector to Teams that sends tasks into a planner. Yeah. So each of those would be a different set of harnesses. Yes. So it'll be harnessing different attributes and aspects.

Speaker 1:

Exchange

Speaker 2:

or teams or

Speaker 1:

So therefore, do you name the harness to describe what's inside the harness? So if I have a folder which has got spreadsheets, PowerPoint, and documents in associated with a project that I'm working on.

Speaker 2:

Yeah. That's a project

Speaker 1:

Then I name the folder. It's in a folder. Yeah. So it's in a folder called project seventy two. Yeah.

Speaker 1:

Does Harness have a name then?

Speaker 2:

Well, you you name the agent, and the agent is the model and the harness combined. So I don't know. Well, you do name a library as well. I mean, you call it either documents, shared documents, published documents, or invoices or whatever. So, yeah, why not?

Speaker 2:

If you think of this in

Speaker 1:

terms of workflows, you would name the environment.

Speaker 2:

Would you create types of harnesses where you would go like this would be a, business process harness. This would be a request harness. This would be a finance harness. This would be an HR harness.

Speaker 1:

But it's a little bit like that content type again. Although the content type does have a boundary. It's a bit more solid. This harness doesn't exist. It's a name.

Speaker 2:

No. It's it's a number of things that you, yeah, conjure up together.

Speaker 1:

Yeah. So, yeah, it's a waste of time really. I guess it's... Okay. Well, well, is, isn't I mean, let's assume that you're doing IT.

Speaker 1:

You're doing IT, so you do a documentation for IT. So they... Yeah. These are all my workflows. These are all my agents.

Speaker 1:

These are all my files. This is how I govern my... I have documentation and everything. So for the agent, I need to describe the harness so that I'm fully encompassing everything around that thing, that automation.

Speaker 2:

You've got content types like procedures, manuals, invoices, whatever. So you can name those. But but they... Yeah. The harness is like a combination of a bunch of things.

Speaker 2:

So that's why

Speaker 1:

It's a technical term for something.

Speaker 2:

Yeah. So that's why I think we could probably categorize them in being like, this is an HR harness. This can only look at HR data with HR sensitivity labels. Only HR people can use it. Only

Speaker 1:

But I don't think it's a real entity. I don't think you go somewhere and say Well, it could be Harness.

Speaker 2:

It could be like an environment in Power Automate where you say...

Speaker 1:

You would be able to open the environment and look at the settings for the environment. I I don't know whether harness is just a descriptive term that we use for a collective set of We're

Speaker 2:

really talking of our ass. Right?

Speaker 1:

Yeah. Yeah. We are. Yeah. But but I've I've never I've never...

Speaker 1:

Okay. I'll ask. Ask.

Speaker 2:

Yes. Ask. Are there types of harnesses? And then she'll open up Amazon and order a bunch.

Speaker 1:

Hey, Holly, when it comes to harnesses, do they... Are they ever physical entity? Can you name them? Or are they just some kind of virtualized way of collecting all your AI parts together?

Speaker 3:

Yeah. Good question. A harness is mostly software, but it's not just a vague idea. It's concrete components that you can point at. So system prompts stored in files, config in JSON or YAML, rag indexes in a vector store, permission groups in intra, logging and monitoring in your dashboards.

Speaker 3:

And sometimes there is a light UI on top, like an admin panel or agent console, but the meat is in those artifacts and controls. For your setup, the harness shows up as SharePoint libraries with strict permissions, intra roles, skill configs, and that certificate control logic you've talked about.

Speaker 1:

Okay. Yes. Let's not give away my IPR. That's fine. Because you're being recorded on the podcast.

Speaker 1:

But I've I've never seen the word, what do you call this harness, in any AI tool or Copilot or anything like that. So whereas I might use the word harness if I was documenting an AI agent to be able to list what's involved in that automation. But I'm guessing you don't have to fill out the word harness or name it. Or you don't use... You don't have a harness template, for example.

Speaker 3:

Exactly. Most platforms don't call it a harness in the UI. It's more of an engineering concept or a way to talk precisely about all the surrounding stuff you build. Some teams formalize it with a repo structure or repo template. Some use an internal checklist or Terraform modules.

Speaker 3:

The label is optional. The discipline's what matters.

Speaker 1:

Alright. Got it.

Speaker 3:

Thank If you wanted to be...

Speaker 1:

I'm gonna go now. Ciao.

Speaker 3:

Okay. Glad you got what you needed.

Speaker 2:

She's so nice.

Speaker 1:

Yeah. She is. Lovely. So, no. So there's no there's no formal description like an authorized model that you need to have stuff at the moment.

Speaker 1:

I mean, might change, of course. Because we do need some

Speaker 2:

Maybe I should coin this, like, harness types of...

Speaker 1:

If I... You're gonna harness this definements....

Speaker 2:

In a

Speaker 1:

book, but you wanted to write a book, there you go. This is the one you've got about a week and a half before somebody else does it.

Speaker 2:

Exactly. Before this episode comes out.

Speaker 1:

Yeah. But but I do like the concept very much that you need a way of collecting the parts associated with this thing that I've built

Speaker 2:

Yeah.

Speaker 1:

Which we call agents. Yeah. And that agent consists of some really smart things that I need to know and store and back up. Yeah. But then I need to find a collective name for everything.

Speaker 1:

But it's it's only... I only use the harness Yeah. But... If have to write it down somewhere. Unless I'm wrong, Zoe and Kevin, and then I'm sure you can correct us.

Speaker 2:

But I mean, governance kind of is the same thing. It's also like a collection of a bunch of things.

Speaker 1:

Correct.

Speaker 2:

It's like rules about security, about setup, about who gets to do what. And... Yeah. Maybe.

Speaker 1:

Maybe. So this is... What you're saying is this is AI governance is called the harness, we think.

Speaker 2:

Yeah. Yeah. I think so. Then I

Speaker 1:

go back to my point that that means you need to be able to hang some instructions and rules from it.

Speaker 2:

Yeah. I think so. I mean, if you say to your agent, you need to be concise and you need to, I don't know, always don't use em dash or whatever. That's... Is that part of the harness?

Speaker 1:

Well, I don't know. I think that's part of the guide rails that you put into the skills and all that kind of stuff. So I think harness is a virtual name thing at the moment, and I hope to be proved wrong. Yeah. Okay.

Speaker 1:

I'll dig around

Speaker 2:

a for everything.

Speaker 1:

Yeah. It's a way of saying, yeah, you give it a name, I suppose. But quite honestly, it's just gonna take the name of the agent. Yeah. Because that's the only thing that you do name that is the collective parts of the agents running an API or if it's pointing at a data source or it's ranked to a library or

Speaker 2:

ranked a document. Why don't we have, like, types of agents? I mean, like HR agents or... Well, frontier

Speaker 1:

crew, we've got HR, finance. So we've got all agents for the for the main business in the frontier crew.

Speaker 2:

Okay. And then you got the agent bosses. So that's a category.

Speaker 1:

Potentially. And...

Speaker 2:

Yeah. But, I mean, there's different agents with different capabilities. So I guess they can be categorized, I think.

Speaker 1:

Yeah. I think you might be be right, but but there's no standards around this, is there?

Speaker 2:

Yet. Maybe we should come come up with something. Yeah. Yeah. I don't know.

Speaker 1:

But there's no standards of that To work by this

Speaker 2:

be honest, I haven't done jack shit with AI. I mean, the only thing I do is I talk to my copilot every once in a while. But, I mean, I've never done anything that drives business value for a customer that has to do with AI. Oh, well. To be honest.

Speaker 1:

I'm quite enjoying my current AI project. Yeah. So that is turning out to be some fun. But it's mostly product owner. My role is product owner deciding what the agents will do Yeah.

Speaker 1:

And what their skills will be. So you're playing with

Speaker 2:

the harness.

Speaker 1:

Charge. I'm trying to work out where the harness fits in this. Yeah. So you're right. My my cosmic closer, which is an agent that is going to support salespeople and sales process within an SME, I guess that is the harness.

Speaker 1:

It's a name. It's an agent. It consists of of rags, permissions.

Speaker 2:

Guardrails.

Speaker 1:

Yeah. So there you go. My frontier crew are harnesses. They're harnessed.

Speaker 2:

Yes. Okay.

Speaker 1:

Yeah. Madam Catalyst. She'd look good in a harness.

Speaker 2:

Madam Catalyst. Madam Catalyst. Okay.

Speaker 1:

She's a change agent.

Speaker 2:

Okay.

Speaker 1:

AI adoption and training agent.

Speaker 2:

Okay. Oh, nice. Okay.

Speaker 1:

It's quite cool, actually. So we're looking at doing things like, looking at the way you operate Outlook, and then she'll give you some suggestions on how you can use AI to improve. And then she'll set a meeting up a week later so that you can go and check, and then she'll basically confirm whether or not this is a habit, and then you'll get a new badge. And so we'll then have a company chart all fully automated that will place you about the people that are actually...

Speaker 2:

Okay. So... So gamification. It is a bit. Yes.

Speaker 2:

Okay. Yeah. It's fun. Cool. Cool.

Speaker 2:

Cool. Cool. Cool. Cool. So many people are doing so many cool things.

Speaker 1:

Yeah. My problem is trying to protect these. I was talking to Kevin while we exchanged a message or two. How do you protect your skills? I think I have a way of doing it with PKI certificates, but

Speaker 2:

I've got to put it in the... Skills as in the AI written So

Speaker 1:

if if for example, I sell somebody an agent, I don't wanna sell them an agent. I wanna lease the agent. Yeah. Keep it updated like an application. Yeah.

Speaker 1:

So if they stop Basically, paying like a SaaS application. Yeah. Exactly. If they stop paying me Yeah. Then I wanna be able to say, bugger off.

Speaker 1:

You can't have those skills anymore. So I then need to be able to lock them lock them down. And I think I found a way of doing it. I've just gotta technically... It's it's way beyond my technology, so I need to give it somebody else Yeah.

Speaker 1:

To technically test it. So the product, I think I have designed, just got to I'm having some fun at the moment. Sorry. Okay. So...

Speaker 1:

Yes.

Speaker 2:

Okay. I'm having fun

Speaker 1:

with this. Yeah. We'll get into it, folks. It launches at the end of September. You'll hear more.

Speaker 1:

I'll do a podcast on it at that point.

Speaker 2:

Yeah. Yeah. We we might get a sponsor. Might get a bottle of whiskey out of this.

Speaker 1:

Yes. I'd have to pay for it myself.

Speaker 2:

Good. Oh, well. Yeah. We'll we'll see how that pans out. Yeah.

Speaker 2:

That's cool.

Speaker 1:

Good. Good. Alright. So we were gonna talk about some of the things that are inside that harness.

Speaker 2:

Yes. So

Speaker 1:

that's what I thought we would try and look at and some of the things that are changing because things are starting to settle down a bit in the AI

Speaker 2:

space. No, they're not.

Speaker 1:

Yeah, they are a little bit. Okay. So there's... It's certainly more solid than it was six months ago. People were panicking about it.

Speaker 1:

They were... Didn't know where it was gonna work. They weren't sure what the users were. They were talking about what permissions were doing, blah blah blah blah blah. Mhmm.

Speaker 1:

And I remember South Coast Summit. When was that? Two years ago now? When we were at South Coast Summit?

Speaker 2:

I was so bloody drunk then, so I don't really remember. Yeah.

Speaker 1:

But that period of time, I remember loads of sessions about, is your data secure because AI is going to blow it all apart? And that never really happened. No. Co pilot still respected that permissions, and yes, if you got your permissions wrong, but if you got them wrong, you got them wrong. But I think those things have kind of settled down a little bit now.

Speaker 1:

They've leveled off. You know you either have it done or you don't have it done. Nobody's necessarily panicking about it. The real panicking now is about how big the data center's going to be, which pisses Yeah. Me

Speaker 2:

Yeah. Yeah.

Speaker 1:

Yeah. You know, I mean, these things change out. What's interesting is that I did some numbers. Did I talk about this on the podcast before? Maybe I did.

Speaker 1:

Tell me if I did. But I have this this this thing. A lot of people I talk to, lot of my kids, although they're slowly changing, they say, yeah, AI is a bad thing. It's bad for the environment and blah blah blah. And you keep reading these fairly ignorant things.

Speaker 2:

But humans are bad for the environment.

Speaker 1:

AI data centers.

Speaker 2:

Yeah. Yeah. But humans are even worse for the environment.

Speaker 1:

That is true too. And almonds. And almonds.

Speaker 2:

Almonds are... The almonds take up more water than a data center.

Speaker 1:

Is that right?

Speaker 2:

Yeah. To grow an... To grow one almond, you need to use, like, 29 liters of water for one almond, something like that. Wow. Yeah.

Speaker 2:

Yeah. Yeah.

Speaker 1:

Yeah. Well, the dead centers nowadays, most of them don't put the water back. They... They're in a closed loop. So it's not that they're Which

Speaker 2:

makes sense.

Speaker 1:

And it's fine. The power is an issue. But the power's not really an issue, but it's just gonna take time to settle down. So Yeah. I did some some testing because I wanted to to look at this myself.

Speaker 1:

So I took an example of an image. Mhmm. So I created this image. And and I had AI create the image, and I changed it a couple of times. And then I went, okay.

Speaker 1:

Let's just assume that I get somebody on Fiverr to do that. Yeah. Alright. So I say, hey, I need this image. And they charge me a €100 for it or whatever.

Speaker 1:

Mhmm. But we make an assumption that, you know, they create the image, whether they use AI or not, but they'll tweak it and adjust it. But let's assume they're doing it on their machine Yeah. In their desk, and they do... They spend three hours on it.

Speaker 1:

So they do it in an hour, and then I look at it, and I make a few changes and everything else. So let's assume they do three hours on it. And then they send it back to me by email, and away we go. So if you take the total amount of electrical power that's been used to create that on a PC, and you take the total amount of electrical power that's been used to create that image on AI and three other versions of it, how big a difference is there between the two in terms of percentages?

Speaker 2:

Okay. I would guess that there's a big difference.

Speaker 1:

There is a big difference, but which way around is it? Which one uses more electricity?

Speaker 2:

Think the AI would use way more energy.

Speaker 1:

Is that right? Yeah. The laptop will use 1000% more power than the data center would.

Speaker 2:

Really?

Speaker 1:

Yes. It was not point three watts for the data center to do everything. Up and loading, we included everything in there. And it was over 30 watts

Speaker 2:

Okay.

Speaker 1:

For the laptop to be on for three hours doing its stuff. And it's not difficult to calculate. Okay. Alright. So where does that take us?

Speaker 1:

And I know we're now talking about something completely different, but this is just something that's in the back of my head. It's gonna take us to the point where we're actually just centralizing our power usage. Mhmm. So instead of people working from home doing their stuff, it's gonna get done centrally. Microsoft...

Speaker 2:

Oh, you mean like a server?

Speaker 1:

Like a... Yeah. Exactly. Microsoft, already issued, Windows 11 Mhmm. For AI.

Speaker 1:

No applications on it. Just a silly operating system that just talked to AI and everything was done in AI. Yeah. But all the Windows 11 users went, no, don't do that. Blah blah blah blah blah blah blah blah Alright.

Speaker 1:

But if you take this logically where it's going, so I have created this week well over 300 pages of documentation. Okay. Now this week and last week. Alright? So creating a bunch of stuff for for processes and policies and everything else.

Speaker 1:

And I've not used Word once other than to open it up and make a few number of changes. Okay. Cool. Never once. Yep.

Speaker 1:

And if you think logically, that's the way forward.

Speaker 2:

Yeah. Alright. Remember that we had this conversation a year ago where I said, we're not gonna use Word, Excel, PowerPoint

Speaker 1:

And I said, no. That is true. Yeah. Yeah. You're right.

Speaker 1:

You were way ahead of the time. You see? There's You

Speaker 2:

should have written a book or

Speaker 1:

you missed it. Damn. Logically, that's where it's gonna go. You'll go to your telephone, you'll just blah blah blah blah blah, and you'll create a word just say what what you want. Alright?

Speaker 1:

And same with Excel. So consequently, you're not gonna need the power on your desktop. This bullshit about AI data centers taking so much power, it's almost... That's correct. It's redistributing...

Speaker 1:

It's a cumulative. Yeah. But if you're now gonna be able to have lighter machines, less power, alright, at home Thin books. Then it's it's actually gonna save the world's power. The Nux.

Speaker 1:

Yep. Okay. Stop doing it, people. You're pissing me off every time I see somebody's, you know, oh, we've started up a petition to stop this AI data center being built here on this crappy piece of film.

Speaker 2:

Oh, yeah.

Speaker 1:

I mean... One of them really made me smile. It was... Know you when you come down from London on the M 25? Yeah.

Speaker 1:

And you get to the bridge over the River Thames? Yeah.

Speaker 2:

The Dartford Bridge. The Dartford Bridge.

Speaker 1:

Coming from the North, you go to Dartford Bridge. Yeah. There's a services at the... Just at the roundabout before you get there. You have to pull off and go round and back.

Speaker 1:

Yeah. It doesn't matter. It's near Lakeside Shopping Centre. Okay. Okay.

Speaker 1:

And behind that, there's a speedway track, an old fashioned banger racing track.

Speaker 2:

Oh.

Speaker 1:

Okay. On a piece of gravel pits and everything else. Well, Google apparently, they're gonna build a data center there in this crappy old piece of banger track racing thing. It's not a pop circuit. It's just, you know.

Speaker 2:

Yeah.

Speaker 1:

Yeah. You know what they're like. Yeah. And people are actually, no. No.

Speaker 1:

We can't have a data center there. That's ridiculous. There's no houses anywhere near it. Alright? But apparently, there's three butterflies and a pansy.

Speaker 1:

Alright? Oh, yeah. Yeah.

Speaker 2:

Yeah. So... Yeah. Like some kind of weird bat lives there. Yeah.

Speaker 2:

Mhmm.

Speaker 1:

Yeah. So I I just... It doesn't... It doesn't make me smile. But, yeah, all this stuff about we don't need AI and AI should be owned by the people.

Speaker 1:

Yeah. Because that's the way the first industrial revolution did. People went on strike, and then they got their way. No. They changed their work conditions, but they still didn't get to own the factories, you know.

Speaker 1:

Yeah. It's it's interesting. And everybody's using AI.

Speaker 2:

Yeah. Yeah. Yeah. Absolutely. Absolutely.

Speaker 2:

Yeah.

Speaker 1:

So there we go. Let's go on moving back there. So, yeah, we were talking about why things are changing. So one of the things was permission models, I think, is the one we talked about earlier.

Speaker 2:

Yeah. Yeah. Yeah. You said something interesting. Because, I mean

Speaker 1:

You're laid on that sofa, all comfy, aren't you? Super comfortable. All you're doing is I'm just... I'll just throw a question at him in your talk,

Speaker 2:

and I can

Speaker 1:

just go to sleep here. But that's all good.

Speaker 2:

I like it a lot. Yeah. No. No. Because we...

Speaker 2:

We've already talked about the fact, like, for example, with intranets that we need to start designing for AI to look at it instead of designing it for human eyeballs or at least for the two of them. But you said something interesting earlier as well.

Speaker 1:

Yeah. So one of the things that I'm gonna talk about in, it's a group actually. Nook and I are gonna talk about, it's a combined session, Is your permission strategy for your SharePoint environments, for example?

Speaker 2:

Yeah.

Speaker 1:

Now what's interesting enough is that your permissions for your Copilot is all covered. Alright? You know, we know that Copilot will never let you get to information you're not supposed to

Speaker 2:

Yeah.

Speaker 1:

Even if it's running under a general account

Speaker 2:

Yeah.

Speaker 1:

On Foundry or something like that. So all of that is kind of covered. And so the the real question then comes back to what if the permissions are wrong in the first place? Yeah. What It's what everyone is

Speaker 2:

afraid of.

Speaker 1:

Yeah. So one of things we'll be talking about, is that you need to not just try and fix your existing permissions, you need a different strategic permissions model. Okay? Yeah. And we're also gonna talk about...

Speaker 1:

So if you're going to be in to Greb, folks, we're gonna talk about how you can get AI to fix your permission model. So Oh, okay. We're gonna talk about how you can get it to look at what it seems to be doing and keeps it updated and stuff like that. So Okay. And we're gonna use AI to actually try and fix your life cycles as well.

Speaker 1:

So we're gonna talk about a bunch of stuff like that.

Speaker 2:

So is that with the m three six five admin agent?

Speaker 1:

Or... Some of those, but we're actually gonna, talk about how you can create a governance document, and then you can ground AI on that governance document. And AI will then point to a SharePoint site and then go, okay, then this is where you're breaking governance or not breaking governance. So your document life cycle is too long. It's too old.

Speaker 1:

It's too neat. Why are you looking at me like that? It's fucking brilliant.

Speaker 2:

It's a fucking brilliant idea. A few weeks ago, I was thinking about new sessions for next year. And one of the sessions that I came up with is can AI be my m three six five administrator? Yes. So indeed.

Speaker 2:

So create a governance document and then just point AI to your governance document and say, do whatever you want, but these are the rules and you can't break these.

Speaker 1:

Yeah. Yeah. So that's what we would think.

Speaker 2:

That is awesome.

Speaker 1:

So anyway, we we kind of play of looking at that. And again, that we talked earlier about frontier crew. That's exactly how we we're gonna be able to create AI as a service. So go... So going back to that.

Speaker 1:

So... But... So the biggest difference in your security permission security model is that you don't focus on people, you focus on the content. Yeah. Alright?

Speaker 1:

So the content really needs to be the thing that I secure. So your permissions gotta go with RBAC. You don't have a choice. You really need to go RBAC, and then that will get Role based. Yep.

Speaker 1:

And then all the other stuff like labels and, site labels and those kinds of permissions, then that that's the kind of strategy you need to move towards. Yeah. And I think I haven't yet thought about this subs. This is what the next thing I'm gonna be looking at probably the next two or three days. Is is it easier to do a completely new permissions model?

Speaker 1:

From the ground up? Yeah. From the ground up based upon your site structure and the content in those sites, or is it better to actually try and adjust your permission model that you've already got in place? I do not know yet. I haven't got that.

Speaker 1:

But what I do know for certain is that this is now about content more than it's ever been about content, not about the people.

Speaker 2:

So when you've got your HR procedures and you say HR should be having right access to this, how does that change in your RBAC model?

Speaker 1:

The way... I I don't know. As I said, I've got to kind of work this out. But the way I think it's going to be, because we're gonna be using AI, I think the way forward... And please folks, by all means, come and shout at me, but I'm just letting you know this is not fixed in my head yet.

Speaker 1:

But I think that, whereas at the moment we have five sensitivity labels. Yeah? But what if we had 20 labels? Mhmm. And so that identifies HR content and finance content and everything else.

Speaker 2:

Yeah. Yeah.

Speaker 1:

And so you can use AI to automatically apply the label. Yeah. And then you can use the security groups inside

Speaker 2:

The label.

Speaker 1:

Labels to decide who goes in and who doesn't.

Speaker 2:

Okay.

Speaker 1:

And so you should be able to come up with a simple... Well, I would get people started, you should be able to come up with a relatively simple role based stuff where you go, okay, you're in IT, for example. This is your IT stuff. Yeah. It doesn't matter which set point site it's on.

Speaker 1:

It's about how the label is labeled. And then you can matrix it with management as well. So I think so. That I haven't worked this through fully, but I do believe our permissions model needs to change and to focus on content.

Speaker 2:

Yeah. Okay.

Speaker 1:

Because that sensitivity label will then decide which content is allowed into the AI model and which isn't. Yeah. So not only that, you're also making sure the stuff is excluded from something so you have a better way of protecting it. So your sensitivity label, you can say, do not allow this to be put into

Speaker 2:

Yeah. So, like, this is still a working version or this is a published version? No. You can

Speaker 1:

do that, of course. But, but actually, within the sensitive label setting itself and DLP, you can say, do not allow anything with a sensitive label to be added to an LLM, co pilot LLM. Yeah. So the whole strategy becomes around where this content can be used, not who can see this content. Because people won't even bother going looking for that document anymore.

Speaker 1:

They're just gonna ask an AI question. Yeah. Yeah. Exactly. Yeah.

Speaker 1:

So I think that there's a... So our opening three or four slides are all about your security permission model inside your harness is gonna change. Now this is where I found it quite interesting. I know you're in one of your listening moods today, and I get it, and it's okay, but I'm

Speaker 2:

on I'm a

Speaker 1:

thinking. Yeah. I'm on a

Speaker 2:

roll.

Speaker 1:

Yeah. The harness itself is gonna have a set of policies associated with it, I think. It's got no choice. Because if you're if you're gonna be talking about the permissions of the content, when the AI can run, the guidelines for it to be run, what it can actually show people, there needs to be some kind of criteria for a harness. I think.

Speaker 1:

I I I I am gonna dig into this because like you, well, you before me, but we sat and talked about it twenty minutes ago. And I don't know what it is. So I quite like this a lot. But in the frontier crew that I'm putting together, that is my harness. Cosmic closer, salespeople, the harness is defined this way.

Speaker 2:

Yeah. Yeah. And I'm I'm still thinking, like, there pieces of the harness that can be reused? Like, I'm thinking about a knight in a in a harness. They all have a helmet.

Speaker 2:

They all have elbow pads. They all have knee pads. And they could be different, but they could also be the same.

Speaker 1:

Could be. Could be. But I think it's a bit like governance. Yes. Your your harness Okay.

Speaker 1:

I guess you'll apply something to... Some... You'll apply a harness, or I'm gonna copy that harness. I don't know whether that's gonna be. Yeah.

Speaker 1:

I don't know whether it's an entity. That's the thing. I don't think... I think it's just a conceptual thing at the moment, but maybe that will change.

Speaker 2:

It's a little bit like a puzzle. You've got all kinds of pieces and you piece them all together and that combination or maybe more like a dish that you make in the kitchen. You got a bunch of ingredients, you pull that all together, and you got a soup. Yeah.

Speaker 1:

I think you're trying to grab onto something and give it a form. Yeah. And I think that's what we're kind of saying is I don't know whether there is a form or not. Yeah. So I think that...

Speaker 1:

I don't think it's the recipe. I think it's the name. Hey. Let's make spaghetti bolognese tonight. But that spaghetti bolognese can be made different ways by different people.

Speaker 1:

Yeah. Like, I always put orange juice in my spaghetti sauce. Just to just to to... Heresy. Exactly.

Speaker 1:

And not many people do. Yeah. Alright. But it's still spaghetti bolognese. Yeah.

Speaker 1:

So I I I think it's, yeah, it's a great one. Maybe we'll have an answer later. Yeah. Maybe the... Our new sponsor for the next couple of three weeks will have his AI Wiki will be able to tell us all about it.

Speaker 2:

We'll have a definite a definitive answer. Yeah.

Speaker 1:

I want to drink some whiskey.

Speaker 2:

Yes, let's. Yes,

Speaker 1:

we've got an interesting whiskey. So there's a story behind this why Marijn said I had to get off the sofa, the lazy bugger, and he gets off this sofa like I do, and he's twenty years younger than I am. Alright. So this whiskey is called the Chivas 100, but it's not from Chivas. It's from the Chivas brothers.

Speaker 1:

Is that right?

Speaker 2:

Yes.

Speaker 1:

The century... Yeah. The century malt, but it's the Chivas brothers. Yes. So it's an independent bottler.

Speaker 1:

And if somebody put together in the millennium, a whiskey that is created from a 100 single malts. Okay? In one form or another. And what happened was a friend of mine who lives down the road, we get together quite often in my bar upstairs and we taste a few whiskies. He has actually access to a company that that bottles whiskey, so every so often he comes back with a plastic bottle, knows nothing about it.

Speaker 1:

But he also does a lot of refitting, and he was refitting a bar, and the barman said, do you drink whiskey? He went, yeah, I drink whiskey. Yeah. I'm not gonna drink whiskey. I'm only selling beer, and this is what I this is what was in the bar already.

Speaker 1:

So there was five bottles of whiskey. That's what I've got

Speaker 2:

left over.

Speaker 1:

Yeah. Yeah. Yeah. Exactly. And this was one of them.

Speaker 1:

So this was issued for the millennium. It was a celebration of Scottish whiskey. Mhmm. And they basically took a 100 malt whiskies and blended them together.

Speaker 2:

Yeah.

Speaker 1:

But I think, from what I can understand, they did it at the wash level. So they they took it before it went into a barrel, blended it, then put it into barrels. Still interesting. Yeah. Yeah.

Speaker 1:

No. It's interesting. Yeah. So when I tasted this, I went, oh, this is interesting. So I said to Dave, hey, can I steal that?

Speaker 1:

Because that's an interesting whiskey. There was only 2,000 bottles or something. I mean, I did an AI run on the label and that's where I can't remember the exact number. But... Yeah.

Speaker 1:

So it's the Schivers Brothers 100 malts, it's called. Okay. That smells like a blend.

Speaker 2:

It definitely smells like a blend. Like, there's no left or right or... Yeah. Exactly. A little bit of rye.

Speaker 1:

I think a bit of rye and also I get citrus as well, but very faintly. Yeah. Exactly. So it's not wasted. Not like some malt.

Speaker 1:

Some blends we can sit there and go, yes. Yes. I smell something.

Speaker 2:

Pineapple. Yeah, exactly. Oh, no. That was the Harness safe word.

Speaker 1:

Yes. The Harness safe word. I was surprised about that. This took nearly an hour to get a safe word out of it. Give Yeah.

Speaker 1:

It a taste, my friend. It's interesting. I also get toasted orange after a while. Now, I think this goes into the category. It's not bad.

Speaker 1:

But? It's not it's not bad drink.

Speaker 2:

It's not a bad drink. It's well rounded. Yep. It could have been more oily. I'm missing that mouthfeel, although it kinda come...

Speaker 2:

The finish the finish kinda comes back.

Speaker 1:

It has got very few legs.

Speaker 2:

Yeah. It's not very pronounced. No. It's well rounded, but well rounded without without a lot of depth.

Speaker 1:

But the finish is nice. That's where that toasted orange comes out. Exactly. Yeah. And you get a bit of nose after the first.

Speaker 2:

It is it is an interesting finish. Yeah. But it's it's well rounded, but it's it's very drinkable. Very. Very, very, very drinkable.

Speaker 2:

But it... Yeah. It's not pronounced. It doesn't have any depth.

Speaker 1:

No. But how many blends do you taste that have no finish whatsoever? All all have, like This is not doing badly.

Speaker 2:

Yeah. Exactly.

Speaker 1:

But I think because if I I might remember rightly, as I said, I remember one Sunday afternoon, we were clear... We were finished off with a couple of these off bottles, and this one came out. I went, that's interesting. Yeah. And I I scanned it and spent the the couple of some minutes looking at where it came from.

Speaker 1:

It was really intriguing. And I'm fairly certain that they blend them together at the wash. Yeah. And so then they put it into One bar...

Speaker 2:

A couple barrels. Yeah. Yeah.

Speaker 1:

Yeah. So that's an interesting... So in theory, it makes it still a single malt, but they come from different maltings. Yeah. It's not it's not legitimately.

Speaker 1:

Yeah. I get that. But it's an interesting experiment.

Speaker 2:

It it grew up together. They grew up together. Yeah. Something like

Speaker 1:

breathed the same air.

Speaker 2:

Yeah. Got visited by same angels. Yeah. Yeah. It's I like it.

Speaker 1:

I have to say.

Speaker 2:

It's it's it's a drinker's drink. Mhmm. Yeah. It's it's okay. But it's not specific or it's not...

Speaker 1:

Yeah. No. It's not standing out. No. But if you do see it on the shelves, give it a try.

Speaker 1:

That's for certain. It's definitely got a a checkbox of give it a go. Yeah. See what you think. But I don't

Speaker 2:

think you'll find it.

Speaker 1:

I think it's gonna be quite rare. Yeah. Exactly. Yeah.

Speaker 2:

And a cool bottle. I mean, I like the bottle, the nice label, good story. Yeah. Yeah.

Speaker 1:

It's... Yeah. Everything a whiskey should be actually.

Speaker 2:

At least... Yeah. Part of it. Yes. Yes.

Speaker 2:

Yes.

Speaker 1:

Oh, well.

Speaker 2:

Well, the question is if if they would let this at a... At the cask strength, for example, would that be... How different would that be? They would get this at cask strength. I think I did put a drop of

Speaker 1:

water into it last time, but I can't remember.

Speaker 2:

Be a way if they watered down whiskey to 40%, for example, is there a way to get that water back out and then get it back to 46 or 50%?

Speaker 1:

Yes. You need to redistill it. The water would boil off before the alcohol.

Speaker 2:

Okay. So, basically, I need to put this in a pan, put the pan on the stove, let it boil for a little bit until the water separates.

Speaker 1:

Yeah. No. Because if you put brandy into it But then

Speaker 2:

I'll go blind

Speaker 1:

because The... No. It has to be in a sealed container. So that's why I have the...

Speaker 2:

So I need to put a little pan. So

Speaker 1:

you you don't... When you boil the mash, the steam doesn't escape out the top of the thing. No. It recondenses back again. So you'd have to recondense it.

Speaker 1:

Yeah. And then... Then you might get it back up again.

Speaker 2:

Let's not. Let's not

Speaker 1:

do it. No.

Speaker 2:

Sounds too difficult.

Speaker 1:

No. Well, they also lose a a large amount of the fluid itself. Yeah. You'd need a lot. Yeah.

Speaker 1:

Okay. Not one of your best ideas.

Speaker 2:

No. No. Yeah.

Speaker 1:

But an interesting one.

Speaker 2:

It's an interesting one. Yeah. Exactly. Yeah. You can would the barrel be the harness for the whiskey?

Speaker 2:

Part of the harness.

Speaker 1:

Okay. Or am I taking this Two too words. Yes. Fuck it. We're not going down that line.

Speaker 2:

No. I mean, if you look at whiskey, I mean, you've got the barrel, you got the barrel that you finish it in, you got the the bottle, you got the brand, you got the label.

Speaker 1:

It's a really good question. There there is a harness alignment with whiskey. I'm not being funny. It's actually a serious one. Okay.

Speaker 1:

It's the name of the distillery. Yeah. That's your harness. It may not be on the bottle. It have nothing to do with the process, but everything is encompassed in that distillery.

Speaker 1:

The fermentation, the boiling, the wash, the distillation, the bottling, the storing, and everything else, but your whiskey may not actually have the name of the distillery. But that distillery is an entity within the process, but it's just unimportant. It's the collective. Yeah. Well, fuck.

Speaker 1:

I like this. There you go, Kevin. I've worked it out, mate. A harness is a distillery. When I was looking for distilleries to go to when we're at Scotland in a few times for a visit, I came across five distilleries we just couldn't go to.

Speaker 1:

Why not? Because they were part of the what is the really, really, really big Diaggio? Yeah. There were... There's one just around the corner from where our barrels are stored.

Speaker 1:

A huge Diaggio distillery that you can't go visit in because it's just full production creating large scale whiskers. And there's another distillery which is a small distillery. Okay. And it's a new one that's been created. I thought, that's that's on the road between where we're going.

Speaker 1:

But you can't go visiting there, but they do have whiskies. But the name of the distillery, it's got nothing to do with the bottles of whiskey that comes out. Interesting. Okay. Do you know?

Speaker 2:

But they've got the same thing at at the Etrador. You can't visit it because it's too small.

Speaker 1:

The Etrador Distillery makes Etrador whiskey in

Speaker 2:

this case. Yeah. But

Speaker 1:

there were a lot of distilleries that are called Jack Jones that makes whiskey called I'll Get You Pissed. Arabella... Yeah. Arabella's Dram. Yeah.

Speaker 1:

So so, yes, there was quite... There's there's one distillery, all modern and everything else, and you look at the bottles and go, I know those bottles, but I've never heard of that distillery. Okay. So so that's the harness. Okay.

Speaker 1:

I like that a lot. Okay. Alright. Next podcast. The whiskey process and AI process and see whether we can actually align them together.

Speaker 1:

Now the upside is that we forget these questions, so we never remember to do it anyway.

Speaker 2:

And we never listen to the podcast anyway.

Speaker 1:

So... I have to when I mix them down, but I won't remember that as a subject. No. So it'd be like tonight. We sat there tasting some whiskey from my collections, some nice ones, and then we went, what we're gonna talk about?

Speaker 1:

Yeah. As we do. As we do. That's it then, I think.

Speaker 2:

I think so. I think that's it. It feels like we've been rabbiting on for quite a while.

Speaker 1:

We're we're just over the hour. Okay. Do we have a title for this yet?

Speaker 2:

Harnessing. Harnessing. Yeah. Whiskey Harness. Yeah.

Speaker 2:

Tying down. Tied tied down. Marijn's tied down in AI.

Speaker 1:

Oh, yeah.

Speaker 2:

I mean, the the the cover image is gonna be interesting. And on

Speaker 1:

that note, bye, guys. Thanks for listening.